Abstract
Smart contracts are computer programs running on the blockchain, and their security issues have received widespread attentions. However, existing vulnerability detection techniques such as static analysis, symbol execution, and fuzz testing could be unable to expose new types of vulnerabilities in time because they rely on extracting vulnerability features manually. On the other hand, vulnerability detection techniques based on machine learning do not require experts to define features manually and can learn vulnerability patterns automatically. In this paper, we propose a vulnerability detection technique for smart contracts, which extracts smart contract features based on the abstract syntax tree (AST) and trains the model to detect smart contract vulnerabilities. Experiment results show that the model achieves an average accuracy of 98.7% and F1-value of 0.885 in detecting three kinds of security vulnerabilities.
Published Version
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call